Journal of Hydrology 366 (2009) 46–54
Contents lists available at ScienceDirect
Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol
Adapting the Water Erosion Prediction Project (WEPP) model for forest applications
Shuhui Dun a,*, Joan Q. Wu a, William J. Elliot b, Peter R. Robichaud b, Dennis C. Flanagan c, James R. Frankenberger c, Robert E. Brown b, Arthur C. Xu d a Washington State University, Department of Biological Systems Engineering, P.O. Box 646120, Pullman, WA 99164, USA b US Department of Agriculture, Forest Service, Rocky Mountain Research Station, Moscow, ID 83843, USA c US Department of Agriculture, Agricultural Research Service (USDA-ARS), National Soil Erosion Research Laboratory, West Lafayette, IN 47907, USA d Tongji University, Department of Geotechnical Engineering, Shanghai 200092, China article info summary
Article history: There has been an increasing public concern over forest stream pollution by excessive sedimentation due Received 11 July 2008 to natural or human disturbances. Adequate erosion simulation tools are needed for sound management Received in revised form 6 December 2008 of forest resources. The Water Erosion Prediction Project (WEPP) watershed model has proved useful in Accepted 12 December 2008 forest applications where Hortonian flow is the major form of runoff, such as modeling erosion from roads, harvested units, and burned areas by wildfire or prescribed fire. Nevertheless, when used for mod- eling water flow and sediment discharge from natural forest watersheds where subsurface flow is dom- Keywords: inant, WEPP (v2004.7) underestimates these quantities, in particular, the water flow at the watershed Forest watershed outlet. Surface runoff Subsurface lateral flow The main goal of this study was to improve the WEPP v2004.7 so that it can be applied to adequately Soil erosion simulate forest watershed hydrology and erosion. The specific objectives were to modify WEPP v2004.7 Hydrologic modeling algorithms and subroutines that improperly represent forest subsurface hydrologic processes; and, to WEPP assess the performance of the modified model by applying it to a research forest watershed in the Pacific Northwest, USA. Changes were made in WEPP v2004.7 to better model percolation of soil water and subsurface lateral flow. The modified model, WEPP v2008.9, was applied to the Hermada watershed located in the Boise National Forest, in southern Idaho, USA. The results from v2008.9 and v2004.7 as well as the field obser- vations were compared. For the period of 1995–2000, average annual precipitation for the study area was 954 mm. Simulated annual watershed discharge was negligible using WEPP v2004.7, and was 262 mm using WEPP v2008.9, agreeable with field-observed 275 mm. Ó 2008 Elsevier B.V. All rights reserved.
Introduction however, subsurface lateral flow and channel flow are predomi- nant (Luce, 1995). When used under such forest conditions, WEPP Many areas of the world depend on forest watersheds as (v2004.7) underestimates subsurface lateral flow and discharge at sources of high-quality surface water for domestic supply, indus- the watershed outlet (Elliot et al., 1995). trial use, and agricultural production (Dissmeyer, 2000). There is WEPP was intended to be applied to agriculture, rangelands and increasing public concern over forest stream pollution by excessive forests (Foster and Lane, 1987). Runoff generation was by the sedimentation resulting from forest management activities. Ade- mechanism of rainfall-excess described by the modified Green- quate erosion simulation tools are needed for sound forest re- Ampt infiltration model (Mein and Larson, 1973; Chu, 1978). Re- source management. The Water Erosion Prediction Project cently, scientists have increasingly realized that rainfall-excess is (WEPP) model (Flanagan et al., 2001), a physically-based erosion not the only runoff-generation mechanism that influences forest prediction software program developed by the US Department of hydrology and erosion (Elliot et al., 1996; Covert, 2003). Forest Agriculture (USDA), has proved useful in areas where Hortonian lands are exemplified by steep slopes, and young, shallow, and flow dominates, e.g., in forest applications of modeling erosion coarse-grained soils, differing markedly from typical agricultural from insloped or outsloped roads, or harvested or burned areas lands. In addition, forest canopy and residue covers differ from by wildfire or prescribed fire (Elliot et al., 1999; Elliot and Tysdal, those in crop- and rangelands causing the rates and combinations 1999; Elliot, 2004; Robichaud et al., 2007). In most natural forests, of individual hydrologic processes to differ (Luce, 1995). Fig. 1 illustrates the differences in major characteristics of hydrologic * Corresponding author. Tel.: +1 509 335 5996; fax: +1 509 335 2722. processes in agricultural and forest settings. WEPP is a reasonable E-mail address: [email protected] (S. Dun). tool in quantifying runoff and erosion from typical agricultural
0022-1694/$ - see front matter Ó 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.jhydrol.2008.12.019 S. Dun et al. / Journal of Hydrology 366 (2009) 46–54 47
P P
Tp Tp Es Es R Surface soil Horizon A R Weathering Horizon B layer Bedrock Horizon C R s
Dp Dp (a) Agricultural (b) Forest
Fig. 1. Diagram showing the difference in primary hydrologic processes between typical (a) agricultural soil profile and (b) forest setting. The size of the arrows reflects the relative magnitude or rate of the individual processes: P, precipitation; Tp, plant transpiration; Es, soil evaporation; R, surface runoff; Rs, subsurface lateral flow, and ; Dp, percolation through bottom of soil profile.
fields (Laflen et al., 2004). For forest watershed applications, how- The erosion routines estimate interrill and rill erosion as well as ever, the model needs to be modified to properly represent the sediment transport in channels. The channel component consists hydrologic processes involved. Covert et al. (2005), in an applica- of channel hydrology and erosion. Channel hydrology routines gen- tion of WEPP for simulating runoff and erosion on disturbed forest erate hydrographs by combining channel runoff with the surface watersheds, emphasized the need for adequately representing lat- runoff from upstream watershed elements, i.e., hillslopes, channels eral flow processes in WEPP. or impoundments. The channel erosion routines simulate soil The main purpose of this study was to improve the WEPP detachment from channel bed and bank due to excess flow shear (v2004.7) watershed model such that it can be used to simulate stress through downcutting and widening as well as sediment and predict forest watershed hydrology and erosion. The specific transport and deposition. objectives were to identify and improve WEPP (v2004.7) subrou- tines for subsurface lateral flow process; and to assess the perfor- Simulation of subsurface water flow in WEPP mance of the modified model by applying it to a typical forested watershed in the US Pacific Northwest. WEPP conceptualizes a hillslope as a rectangular strip with a representative slope profile and multiple OFEs (Cochrane and Flan- Method agan, 2003). Rainfall excess, in intervals of minutes, is calculated as the difference between rainfall rate and infiltration estimated Model description using a modified Green-Ampt- Mein–Larson equation (Mein and Larson, 1973; Chu, 1978), with rainfall interception in the canopy WEPP partitions a watershed into hillslopes and a channel net- and residue as well as surface depression storage taken into con- work that includes channel segments and impoundments. Over- sideration. Overland flow rate on an OFE is determined from the land flow from hillslopes feeds into the channel network. A average rainfall excess of the immediately upstream, consecutively hillslope can be further divided into overland-flow elements feeding OFEs weighted by their lengths. Redistribution of infil- (OFEs), within which soil, vegetation, and management conditions trated water, including evapotranspiration (ET), percolation, and are assumed homogeneous. A recently developed geo-spatial inter- subsurface lateral flow, is simulated within each OFE on a daily ba- face, GeoWEPP, allows the use of digital elevation models (DEMs) sis. Transmission of subsurface lateral flow is between two adja- to generate watershed configurations and topographic inputs for cent OFEs. the WEPP model (Renschler, 2003). Important functions and rou- Within an OFE, a soil profile can be divided into layers of 10 cm tines in WEPP are summarized below after Flanagan et al. (1995). for the top two (for better description of surface conditions, e.g., The hillslope component of WEPP simulates the following pro- tillage effect) and 20 cm for the remainder. Soil physical properties cesses: surface hydrology and hydraulics, subsurface hydrology, for each layer, such as saturated hydraulic conductivity (Ks), field vegetation growth and residue decomposition, winter processes, capacity (hfc) and plant wilting point (hwp) (considered as the soil and sediment detachment and transport. The surface hydrology water content h at matric potential of about 30 kPa and 1.5 MPa, routines use information on weather, vegetation and management respectively), are estimated using the soil texture input. WEPP al- practices, and maintain a continuous balance of the soil water on a lows a user to define an effective saturated hydraulic conductivity daily basis. The hydraulics routine performs overland-flow routing for the top two soil layers, which is adjusted for tillage practices based on the solutions to the kinematic wave equations, and ad- and soil consolidation, and is used in the Green-Ampt- Mein–Lar- justs hydraulic properties with changes in soil and vegetation con- son infiltration equation. ditions. The subsurface hydrology routines compute lateral flows WEPP simulates percolation (Qp) when soil water content ex- following Darcy’s law. The vegetation growth and residue decom- ceeds field capacity. The amount is calculated using the unsatu- position routines calculate biomass production and residue rated hydraulic conductivity (estimated from Ks, h, hfc, and hwp) decomposition under common management practices. The winter and available water for percolation in the current (ith) soil layer routines simulate soil frost-thaw, snow accumulation and melt. as well as the degree of saturation of the layer below (i + 1th). 48 S. Dun et al. / Journal of Hydrology 366 (2009) 46–54 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 1 Q p ¼ mmi 1 Siþ1½1 expð Dt=tiÞ ð1aÞ where Rs (m s ) is subsurface lateral flow, Kl (m s ) is the equiva- 1 lent lateral hydraulic conductivity, Sp (m m ) is the average slope mmi ¼ðhi hfciÞdi ð1bÞ gradient of the OFE, Dd (m) is the total thickness of the drainable hiþ1 hwpiþ1 Siþ1 ¼ ð1cÞ layers, L (m) is the slope length of the OFE, di (m) is thickness, /iþ1 hwpiþ1 1 and Kui (m s ) is unsaturated hydraulic conductivity as previously hi hfci defined. ti ¼ ð1dÞ Kui hi In WEPP, subsurface lateral flow from the upland OFE is in- h h 2:655 log fci wpi cluded in the soil water input to the current OFE. Simulation /i hwpi Kui ¼ KsiSi ð1eÞ of daily soil water redistribution follows the order of percola- tion, soil evaporation, subsurface lateral flow, saturation-excess, Where Dt (s) is time interval, and Q (m s 1), S [ ], h (m3 m 3), h p i i fci and plant transpiration; and soil water content is adjusted (m3 m 3), h (m3 m 3), / (m3 m 3), d (m), K (m s 1), K (m s 1), wpi i i si ui after each process. WEPP simulates saturation-excess by com- vv (m), and t (s) are percolation, degree of saturation, soil water i i paring soil water content against the porosity for each layer content, field capacity, wilting point, porosity, soil thickness, satu- bottom to top. The excess water for a layer is added to the rated hydraulic conductivity, unsaturated hydraulic conductivity, layer immediately above. When soil water content in the first available water for percolation, and travel time of percolating water 3 3 layer exceeds its porosity, surface runoff occurs due to satura- of the ith soil layer, respectively. Si+1 [ ], hi+1 (m m ), hwpi+1 3 3 3 3 tion-excess. (m m ), and /i+1 (m m ) are degree of saturation, soil water content, wilting point, and porosity of the i + 1th soil layer, Limitations and modifications of subsurface lateral flow routines of respectively. WEPP Percolation through the last layer is considered as deep percola- tion that leaves the model domain, and is estimated following Eq. Surface runoff is transferred to the channel network in WEPP (1) except that the degree of saturation for the media below this (v2004.7). Subsurface lateral flow, however, was neglected. Such bottom layer is set to zero. simplification may be adequate for agricultural settings with rela- WEPP next estimates evapotranspiration (ET) from the soil. tively uniform and deep soil layers, but can cause underestimates WEPP calculates soil evaporation and plant transpiration sepa- of watershed discharge for steep forested areas. Further, WEPP rately. Potential ET is estimated using the Penman (1963) method (v2004.7) tends to under-predict subsurface lateral flow due to when wind data are available. Soil water is withdrawn when resi- its over-prediction of the deep percolation for two reasons. First, due and plant rainfall interception cannot fulfill the potential ET. WEPP uses user-specified effective saturated hydraulic conductiv- The potential soil evaporation is a fraction of the potential ET based ity for the top two soil layers, but estimates Ks for the remaining on the fraction of uncovered soil. Actual soil evaporation is as- layers using pedo-transfer functions, with a lower limit of sumed to take place only in the top soil layer, and is estimated 8 1 2.0 10 ms (Flanagan and Livingston, 1995). The estimated using the Ritchie’s (1972) model. WEPP assumes that ET is solely Ks is generally greater than that for bedrocks, which potentially from plant transpiration when the plant leaf area index (LAI) ex- leads to overestimated deep percolation and underestimated sub- ceeds three. Potential plant transpiration is calculated as a fraction surface lateral flow (calculated after percolation and soil water of total potential ET using one third of the plant LAI. Water is with- reduction) for most forest settings with shallow soils overlying drawn from the soil layers in the root zone when plant rainfall low-permeability bedrock. Second, WEPP assumes isotropic soil interception is insufficient to satisfy potential plant transpiration. layers. This assumption is inadequate for forest lands where the Water uptake from the soil profile is estimated using the following layering of porous soil and low-permeability bedrock, together equations (Flanagan and Nearing, 1995) with the effect of lateral tree roots, leads to an anisotropic system
with a lateral Ks value greater than the vertical value (Bear, 1972; Ui ¼ Upi hi > hci ð2aÞ Brooks et al., 2004). hi To adapt the model for forest applications, we modified the Ui ¼ Upi hi < hci ð2bÞ hci WEPP soil input files to allow the definition of a ‘‘restrictive” Xi 1 bedrock layer beneath a soil profile with user-specified K for Ep s U ¼ ½1 eðchi=hrzÞ U ð2cÞ pi 1 e c j deep percolation. In addition, the user is allowed to input an j¼1 anisotropy ratio of the soil Ks, and the value is used for the whole soil profile. With the presence of a restrictive layer, 1 1 3 3 3 3 where Ui (m s ), Upi (m s ), hi (m), hi (m m ), and hci (m m ) deep percolation is estimated following Eq. (1) except that are, respectively, actual water uptake, potential water uptake, depth Kui in Eq. (1d) is replaced by the harmonic mean of the of the soil layer, soil water content, and critical soil water content hydraulic conductivities of the bottom soil layer and the below which plant growth is subject to water stress for the ith restrictive layer. layer; hrz (m) is the depth of the root zone, c is a parameter for plant Subsurface lateral flow from a hillslope was added to the chan- uptake distribution; and Ep is potential plant transpiration. nel flow under two conditions: (i) when surface runoff and subsur- In the WEPP model, subsurface lateral flow is simulated when face lateral flow occur simultaneously, and (ii) only subsurface soil water content in a layer exceeds its drainable threshold de- lateral flow occurs. In both cases, we assumed that subsurface lat- fined as field capacity corrected for entrapped air. Subsurface lat- eral flow does not contribute sediment to the stream channels. Un- eral flow is calculated from Darcy’s law using unsaturated der the first condition, surface runoff is presumed to dominate hydraulic conductivity of the draining layer and the average sur- erosion and channel flow processes, and subsurface lateral flow face gradient across the OFE as follows: is simply added to the channel flow by volume without changing hydrograph characteristics and the amount of sediment. Under D R ¼ K S d ð3aÞ the second condition, a uniform flow rate and a 24-h flow duration s l p L were assumed. D ¼ Rd ð3bÞ d i The modifications were made on WEPP v2004.7 and included in RðdiKuiÞ WEPP v2008.9 (accessible to public at http://topsoil.nserl.pur- Kl ¼ ð3cÞ Dd due.edu/nserlweb/weppmain). S. Dun et al. / Journal of Hydrology 366 (2009) 46–54 49
Model application were missing for the spring and summer of 1998. Dew-point tem- perature data were replaced with estimates based on daily maxi- Study site mum and minimum temperatures following Kimball et al. The 9-ha Hermada watershed was chosen as the test watershed (1997). An erroneous wind velocity (or solar radiation) value for for this study. A brief site description, summarized from Covert a specific day was replaced by the average of the wind velocities et al. (2005), is given below. The Hermada watershed is located (or solar radiation data) for the same day in other years. in the Boise National Forest, Idaho (43.87°N and 115.35°W) with The processed climate data are compatible with PRISM-esti- an elevation range of 1760–1880 m and slope gradients of 40– mated (OCS, 2006) data and data observed from the nearby Gra- 60%. Trees, predominantly ponderosa pine and Douglas fir, were ham Guard SNOTEL Station for the same period of time. Fig. 2 harvested in 1992 using a cable-yarding system, and a prescribed shows the comparison of monthly precipitation as an example. fire was ignited on October 17, 1995 for site preparation. The wa- Fig. 3 presents the climate inputs: daily precipitation, temperature, tershed was extensively monitored for discharge using a Parshall solar radiation, and wind velocity, for the WEPP application. flume and sediment with a sediment trap from November 3, 1995 through September 30, 2000. Soil. Soil in the study area is a loamy sand (Typic Cryumbrept) from granitic parent material (Robichaud, 2000). Soils of this type in Ida- WEPP inputs ho generally exceed 1 m in depth and are mostly dry in late sum- The period of field monitoring was used as the simulation time mer (Cooper et al., 1991). Major soil inputs for the WEPP model for this study. The majority of the input data were based on field were based on field and laboratory measurements (Covert, 2003) observation and measurements, while the remaining parameters as shown in Table 2. Effective hydraulic conductivity, a sensitive were from the WEPP User Summary and Technical Documentation parameter for infiltration, was calibrated as 2.5 10 5 ms 1, (Flanagan and Livingston, 1995; Flanagan and Nearing, 1995). slightly higher than the value of 1.8 10 5 ms 1 measured in a field infiltration study using rainfall simulator (Robichaud, 2000). 9 1 Topography and burn severity. The watershed structure and slope A restrictive layer with a calibrated Ks of 9.7 10 ms , which files for the WEPP model were generated from GeoWEPP. The wa- was between the Ks values of weathered and unfractured granite tershed was delineated into one channel section and three single- (Domenico and Schwartz, 1998), was used to represent the bed- OFE hillslopes to the south, north and the west of the channel rock beneath the soil profile. The soil depth was 750 mm based (Table 1). The prescribed fire on October 17, 1995 produced an on field observation. An anisotropy ratio of 25 for the soil profile overall low-severity burn on the west (aspect 48° from due north) was specified to reflect the difference between horizontal and ver- and north (aspect 170°) slopes while leaving the south (aspect tical hydraulic conductivities for the Pacific Northwest forest wa- 295°) slope and lower channel unburned (Robichaud, 2000).The tershed conditions following Zhang et al. (2008) and Brooks et al. prescribed fire resulted in a low-severity burn following the classi- (2004). The initial soil saturation level was set at 45%, considering fication of Ryan and Noste (1983) in which only a small portion of the effect of the prescribed fire in the late fall of the first year of the litter and duff were burned with little effect on the remaining simulation. This setting reflects the soil conditions after a relatively standing trees. Infiltration capacity was not substantially altered dry year of 1994 (OCS, 2006) based on the results from a prelimin- based on the results of the field study of Robichaud (2000). ary WEPP run. The depths to a non-erodible layer in mid-channel and along the side of the channel were set to 0.05 m and 0.01 m, Climate. Field-observed precipitation data for the Hermada wa- respectively, consistent with field conditions. tershed contained two sets of measurements: one by a tipping- bucket rain gage in one-min intervals, and the other by a weigh- Management. A perennial setting was used to represent the regen- ing-bucket gage in 15-min intervals. The weighing-bucket gage erating forest. Management inputs were either from field investi- was equipped with Alter-type shields for wind, more suitable for gation or the literature. Initial ground cover, including grasses, catching snow in the winter. In addition to precipitation, an on-site duff and debris, was 95% for the unburned condition and 90% for weather station recorded temperature, relative humidity, solar the burned condition based on field measurements. Major param- radiation, wind velocity, and wind direction. eters for vegetation growth were taken from the auxiliary database Precipitation data from the two gages were examined and com- of the WEPP model for five-year-old trees (Elliot and Hall, 1997). bined to develop daily precipitation data (amount, duration, time Other parameters were adjusted to represent the tree growth pro- to peak, and peak intensity). Generally, data from the gage that cesses for the study area and to attain agreement between the sim- exhibited greater consistency and caught more precipitation was ulated and observed ground cover over time. These parameters used. include the initial canopy cover, senescence date and length, frac- Additionally, faulty data due to equipment malfunction were corrected. Small gaps of precipitation and daily maximum and minimum temperature (roughly 6% of the total data) were filled with the data for the same period from the closest Snowpack 500 WEPP input Telemetry (SNOTEL) site, the Graham Guard Station (43.95°N, 400 PRISM estimation Graham Guard Station observation 115.27°W, 1734 m a.s.l., 11 km to the northeast) in Idaho (USDA, 300 2006). Dew-point temperature, solar radiation and wind velocity 200 100 0 Precipitation (mm) Table 1 Configuration of the Hermada watershed in WEPP simulations. OJAJOJAJOJAJOJAJOJAJO 1996 1997 1998 1999 2000 Hillslope West North South Channel Length, m 240 242 129 120 Fig. 2. Comparison of monthly precipitation for the Hermada watershed. The thick Width, m 142 175 175 1 gray line represents monthly sum of WEPP daily inputs from observed data, the Average slope, m m 1 0.467 0.476 0.399 0.266 dotted line shows PRISM monthly spatial interpolations, and the solid line Area, m2 34,080 42,350 22,575 120 represents the monthly sum of SNOTEL daily observations at the Graham Guard Station. 50 S. Dun et al. / Journal of Hydrology 366 (2009) 46–54
70 a 60
50
40
30
20
Precipitation (mm) 10
0 60 Maximum b Minimum 40 Dew-point C) o
20
0
-20 Temperature ( Temperature
-40 35 c
) 30 -2 25
20
15
10
Radiation (MJ m 5
0 3 d ) -1
2
1 Wind Velocity (m s 0 OJAJOJAJOJAJOJAJOJAJO 1996 1997 1998 1999 2000
Fig. 3. Observed Hermada watershed daily weather inputs used in WEPP, (a) precipitation, (b) maximum, minimum temperature and dew-point temperature, (c) solar radiation, and (d) daily average wind speed.
tions of canopy and biomass remaining after senescence, and Table 2 decomposition constant for above-ground biomass (Table 3). Major soil inputs for WEPP applications to the Hermada watershed.
Parameters Unburned Low-burn WEPP runs and model performance assessment severity Model runs were performed with the same inputs using the Soil depth (mm) 750 WEPP v2004.7 and v2008.9, respectively. Model results from both Sand (%) 85 versions were contrasted and compared with field observations. Clay (%) 2 Simulated growth rate of above-ground living biomass was com- Organic matter (% volume) 5 pared with literature data, and modeled residue ground cover Cation exchange capacity (cmol kg 1) 1.5 Rock fragments (% volume) 20 was compared with the field-observed values. Simulated stream- Albedo 0.1 flow and sediment yield were compared with monitoring data at Initial soil saturation (%) 45 the study site. Additionally, nonparametric analyses of variance 4 6 6 Baseline interrill erodibility (kg s m ) 2.7 10 4.0 10 (ANOVA) were performed at a = 0.05 (SAS Institute Inc., 1990) Baseline rill erodibility (s m 1) 1.0 10 5 3.4 10 4 Baseline critical shear (N m 2) 1.6 and model efficiency coefficients (Nash and Sutcliffe, 1970) were Effective hydraulic conductivity of surface soil 2.5 10 5 2.5 10 5 calculated for simulated annual and daily watershed discharge (m s 1) from WEPP v2004.7 and v2008.9, respectively. Nonparametric 9 Saturated hydraulic conductivity of restricted layer 9.7 10 tests were used because of the lack of independence, non-normal- (m s 1) ity, and small population size, typically associated with annual and Soil anisotropy ratio 25 daily streamflow samples. S. Dun et al. / Journal of Hydrology 366 (2009) 46–54 51
Table 3 lated annual average volumetric soil water content varied between Major management inputs for WEPP applications to the Hermada watershed. 0.08 and 0.12, averaging 0.09 for the five-year period. Simulated Parameters Unburned Low-burn subsurface lateral flow from the hillslopes was negligible, account- severity ing for 0.15% of annual precipitation on average, as a result of the Initial ground cover (%) 95 90 inadequate simulation of subsurface lateral flow by WEPP v2004.7. Date to reach senescence (Julian day) 300 300 Because there was no predicted surface runoff from hillslopes and Period over which senescence occurs (days) 90 90 because subsurface lateral flow was not passed to the channel, Fraction of canopy remaining after senescence 0.70 0.70 WEPP v2004.7 predicted no watershed discharge. Yet watershed Fraction of biomass remaining after senescence 0.95 0.75 Decomposition constant for above-ground biomass 3.0 10 3 2.0 10 3 discharge was recorded in all monitored years. For the five moni- (kg m 2 d 1) tored years, the observed watershed discharge amounted to 275 mm or 29% of annual precipitation. Hence, WEPP v2004.7 underestimated watershed discharge for the Hermada watershed. Results and discussion Watershed discharge simulated using WEPP v2008.9 improved substantially, with a five-year average of 262 mm accounting for Vegetation cover 28% of average annual precipitation (Table 5). Simulated water flow from the hillslopes was entirely subsurface lateral flow as ob- In comparison to field observations, WEPP (v2008.9) adequately served in the field during this study. Increase in the simulated sub- simulated plant growth and ground cover over time (Fig. 4). The surface lateral flow and thus watershed discharge was a direct ground cover on the unburned south slope remained at 95% during consequence of the reduced simulated soil water percolation due the five-year field monitoring period, and it increased gradually to the restrictive layer specified in the model. Hydraulic conductiv- 14 after the low-severity burn from 90% to full cover on the north ity for an unfractured granitic restrictive layer ranges 10 – 10 1 slope. 10 ms (Domenico and Schwartz, 1998), much lower than The average annual growth rate of the above-ground living bio- the hydraulic conductivity of the soil at the Hermada watershed. mass simulated using v2008.9 was 0.4 kg m 2 for the unburned Consequently, soil water percolation by WEPP was not sensitive slope and 0.3 kg m 2 for the burned slopes (Fig. 4). Herman and to minor changes in the hydraulic conductivity of the restrictive Lavender (1990) state that typical growth rates for Douglas Fir for- layer. ests range from 0.16 to 0.7 kg m 2, depending on climate, soil and Simulated soil water content of hillslopes is typically replen- age of trees. The range of growth rates was for a climate on the US ished over the winter-spring season and starts to decline from Pacific Northwest coast that is generally wetter than the study site. early summer until the dry months of July and August. Fig. 5 shows The estimated growth rates for Hermada watershed appear to be the seasonal change in profile-averaged soil water content for the well within the observed range of growth rates for western US west slope as an example. ET from WEPP v2008.9 was 565 mm forests. on average, accounting for 59% of annual precipitation. Law et al. (2002) presented ET values on evergreen coniferous forest deter- Water balance and sediment yield mined by eddy covariance by various researchers from different study sites. The observed yearly ET varied from 350 to 680 mm Major water balance components for the hillslopes and entire when annual precipitation was between 700 and 1250 mm, the watershed simulated with WEPP v2004.7 and v2008.9 are pre- range of annual precipitation observed at the Hermada watershed sented in Tables 4 and 5, respectively. Annual ET and deep perco- during the study period. Hence, ET simulated from WEPP v2008.9 lation simulated by WEPP v2004.7 essentially accounted for the is compatible with the literature values. Notice the differences in whole water balance. The former averaged 64% and the latter simulated lateral flow and deep percolation among the three hill- 36% of annual precipitation, and no runoff was predicted. Simu- slopes (Table 5). Simulated lateral flow for the south slope was
) 5 100 -2 a b 4 90 3
2 80 1 Predicted
Ground Cover (%) Cover Ground Observed
Living Biomass (kg m Biomass (kg Living 0 70
) 5 100 -2 c d 4 90 3
2 80 1 Ground Cover (%) Cover Ground
Living Biomass (kg m Biomass (kg Living 0 70 OJAJOJAJOJAJOJAJOJAJO OJAJOJAJOJAJOJAJOJAJO 1996 1997 1998 1999 2000 1996 1997 1998 1999 2000
Fig. 4. WEPP v2008.9 simulations: (a) Above-ground living biomass and (b) ground cover simulated for unburned south slope; (c) Above-ground living biomass and (d) ground cover simulated for low-severity burn north slope. 52 S. Dun et al. / Journal of Hydrology 366 (2009) 46–54 )
Table 4 -3 0.6 Annual water balance, in mm, for the Hermada watershed from WEPP v2004.7. m 3 0.5 a b Water year P Slope/watershed QETDp Qs SW 0.4 1995–1996 1106 W 0 497 636 2 87 0.3 N 0 575 545 2 85 S 0 575 551 2 88 0.2 c WS 0 (321) 548 578 0 86 0.1 1996–1997 1200 W 0 621 576 2 82 0.0 N 0 649 546 2 74 OJAJOJAJOJAJOJAJOJAJO Soil Water Soil Water Content (m S 0 647 549 3 79 1996 1997 1998 1999 2000 WS 0 (421) 639 557 0 78 1997–1998 919 W 0 800 123 1 74 Fig. 5. Profile-averaged soil water content for the west slope simulated using WEPP N 0 871 49 0 56 v2008.9. S 0 827 95 1 69 WS 0 (224) 837 85 0 65
1998–1999 809 W 0 460 352 1 59 25 1400 Runoff Precipitation (mm) N 0 463 347 1 55 a 1200 20 Cumulative precipitation S 0 461 350 1 60 Cumulative rain plus snowmelt 1000 WS 0(237) 461 349 0 57 15 800 1999–2000 737 W 0 532 203 1 62 600 10 N 0 576 159 1 49 400 S 0 546 189 1 61 (mm) Runoff 5 200 WS 0 (174) 554 181 0 56 0 0
P = precipitation, Q = surface runoff, ET = evapotranspiration, DP = deep percolation, 25 Qs = subsurface lateral flow, and SW = average soil water storage. b 20 a Water year refers to the period of October 1–September 30 in this study. b W, N, S, and WS refer to west, north and south slopes and watershed, 15 respectively. c Predicted and observed (in parentheses) watershed discharge. 10
Runoff (mm) 5
0
Table 5 ) 0.0010 -1 Annual water balance, in mm, for Hermada watershed from WEPP v2008.9. c 0.0008 a b Year P Slope/watershed QETDp Qs SW 0.0006 1995–1996 1106 W 0 497 231 396 144 N 0 497 231 396 144 0.0004 S 0 492 178 453 132 0.0002 WS 399(321)c 495 219 10 141 Sediment Yield (t ha Sediment 0.0000 1996–1997 1200 W 0 646 144 406 127 OJAJOJAJOJAJOJAJOJAJO N 0 645 144 407 127 1996 1997 1998 1999 2000 S 0 640 109 448 116 WS 403(421) 644 136 11 125 Fig. 6. Hydrograph and sedimentograph for the Hermada watershed. Daily 1997–1998 919 W 0 699 92 126 94 watershed discharge (a) simulated by WEPP v2008.9 and (b) observed, and (c) N 0 699 91 127 94 daily sediment simulated by WEPP v2008.9. S 0 704 64 151 89 WS 125 (224) 701 86 6 93 1998–1999 809 W 0 492 106 219 96 of 1995–1996, as corroborated by field observation. Winter runoff N 0 490 109 219 97 S 0 475 90 252 97 was not recorded between November, 1996 and February, 1997 in WS 218 (237) 488 104 8 97 our study due to a malfunction in the data acquisition device; how- 1999–2000 737 W 0 496 74 167 67 ever, the large simulated winter runoff coincided with the 1996– N 0 496 73 168 68 1997 winter flooding in the western US due to heavy precipitation S 0 493 55 189 64 and warm temperatures (Lott et al., 1997). Overall, simulated and WS 167 (174) 495 69 5 67 observed daily runoff during spring snowmelt seasons for the Her-
P = precipitation, Q = surface runoff, ET = evapotranspiration, DP = deep percolation, mada watershed were agreeable. Good agreement was also found Qs = subsurface lateral flow, and SW = average soil water storage. for simulated and observed annual runoff (Fig. 7). a Water year refers to the period of October 1–September 30 in this study. Sediment yield of individual storm events was monitored dur- b W, N, S, and WS refer to west, north and south slopes and watershed, respectively. ing the study period, but no sediment-producing event, i.e., with 1 c Predicted and observed (in parentheses) watershed discharge. a sediment yield P 0.005 t ha , was observed (Covert, 2003). WEPP v2004.7 simulated no runoff from either the hillslopes or channel and thus no soil loss. WEPP v2008.9 also simulated no soil more than those for the west and north slopes because of its short- loss from hillslopes. WEPP-generated erosion was due to channel er slope length (see Table 1). Greater simulated lateral flow in turn flow with a maximum daily sediment yield less than 0.001 t ha 1 led to less simulated deep percolation for the south slope. (Fig. 6c). WEPP-simulated annual sediment yield varied from For the study area, observed streamflow mostly occurred in the 0.028 t ha 1 in the first year to 0.005 t ha 1 in the fifth year after spring snowmelt season and occasionally in warm winter or wet the prescribed fire (Fig. 7). These values are rather low for recently summer (Fig. 6). WEPP v2008.9 generated substantial winter run- burned areas, but greater than the field-observed values during off for the 1996 and 1997 water years. WEPP-simulated winter those years (Covert, 2003). Annual soil erosion rates after pre- runoff for the 1996 water year corresponded to the warm winter scribed fires can vary from 0.1 t ha 1 for low-severity burns to S. Dun et al. / Journal of Hydrology 366 (2009) 46–54 53
600 0.030
Sediment Yield (t ha (t Sediment Yield difference or finite-element techniques to solve the governing par- Observed runoff Simulated runoff (v2008.9) 0.025 tial differential equations for major hydrological processes, and Observed sediment yield therefore can be more data-demanding and computation-expen- 400 Simulated sediment yield (v2008.9) 0.020 sive. These models do not have options as comprehensive as the 0.015 WEPP model for assessing the effects of different management 200 0.010 practices (e.g., culvert and settling basin) on site-specific hilllslope
Runoff Depth (mm) 0.005 erosion, thus limiting their applicability at forest project scales. -1 ) 0 0.000 1996 1997 1998 1999 2000 Summary and conclusions Water Year Reliable models for simulating water flow and sediment dis- Fig. 7. Annual runoff and sediment yield from the Hermada watershed. charge from forest watersheds are needed for sound forest man- agement. WEPP, a process-based, continuous erosion prediction model, was adapted for forest watershed applications. Specifically, Table 6 Nonparametric ANOVA results and Nash–Sutcliffe model performance coefficients modifications were made in modeling deep percolation of soil from comparing means of observed and WEPP-simulated annual or daily watershed water and subsurface lateral flow. The refined WEPP model has discharge (a = 0.05). the ability to simulate subsurface lateral flow through the use of WEPP version Nonparametric ANOVA Nash–Sutcliffe a restrictive layer and an anisotropy ratio specified by the user. coefficient Further, it is capable of transferring subsurface lateral flow from Annual Daily Annual Daily the hillslopes to watershed channels, and then routing it to the wa- tershed outlet. Compared to WEPP v2004.7, WEPP v2008.9 may be F-value P-value F-value P-value used to more realistically represent hydrologic processes in forest ** ** 2004.7 40.3 0.0002 302.8 <0.0001 10.0 0.17 settings. 2008.9 0.03 0.86 0.35 0.56 0.57 0.45 The comparison of WEPP model results from v2004.7 and ** Significantly different. v2008.9 for the Hermada watershed, a representative, small forest watershed in southern Idaho, showed the improvement of the modified model in simulating subsurface lateral flow from hill- 1 6.0 t ha for high-severity burns (Robichaud and Waldrop, 1994). slopes and daily and annual streamflow. WEPP v2008.9 yielded Sediment yields after fire depend on many factors, such as climate, predominant seasonal subsurface lateral flows, consistent with vegetation, topography, and soil (Swanson, 1981). The low post- field observation. fire erosion rate at the study watershed was likely due to the low For steep mountainous forested watersheds with granitic bed- impact from the low-severity burn on ground cover and its rapid rock underneath shallow soils, such as the Hermada watershed, recovery (Fig. 4b). the hydraulic conductivity of the bedrock is sufficiently low that the small changes in the hydraulic conductivity would not greatly Statistical analysis affect deep seepage; and, lateral flow may oftentimes be the major contributor to observed streamflow. The use of a perennial plant No runoff was generated using WEPP v2004.7. Nonparametric setting in this study reasonably described vegetation regeneration ANOVA results (Table 6) show that both annual and daily mean and ground cover under forest conditions. watershed discharges simulated using WEPP v2004.7 differed sig- nificantly from field-observed values. Runoff results from WEPP Acknowledgments v2008.9 were substantially improved as demonstrated by the non- significant ANOVA test results for both annual and daily values. This study was in part supported by the funding of the USDA The Nash-Sutcliffe coefficient for daily watershed discharge further National Research Initiative (Grant No. 2002-01195) and of the indicates the inadequacy of WEPP 2004.7 ( 0.17) and agreement USDA Forest Service Rocky Mountain Research Station through a between WEPP 2008.9 simulation results and field observations Research Joint Venture Agreement (No. 05-JV-11221665-155). (0.45). We thank S.A. Covert for her suggestions about WEPP input prep- Compared to v2004.7, v2008.9 yielded daily hydrograph and aration, D.K. McCool for the valuable discussions on winter hydro- annual streamflow values that were more agreeable with field logic processes, and C.O. Stöckle for his comments that helped to observations. Yet there are limitations of WEPP for applications improve the technical rigor and editorial clarity of the manuscript. to certain conditions. In the current version of WEPP, deep perco- lation contributes to ground water, which does not interchange References with watershed streamflow. Hence, WEPP cannot be used for areas where surface- and ground-water interaction is important. Bear, J., 1972. Dynamics of Fluids in Porous Media. Dover Publications, Inc., New WEPP is a network-based, hydrologic-unit model. It discretizes York. pp. 124. Brooks, E.S., Boll, J., McDaniel, P.A., 2004. A hillslope-scale experiment to measure a watershed into hillslopes and a channel network. A hillslope can lateral saturated hydraulic conductivity. Water Resour. Res. 40, W04208. be divided into the basic simulation units of OFEs representing dif- doi:10.1029/2003WR002858. ferent vegetation, soil and topography conditions. Consequently, Chu, S.T., 1978. Infiltration during an unsteady rain. Water Resour. Res. 14, 461– 466. WEPP can simulate saturation-excess runoff due to changes in hill- Cochrane, T.A., Flanagan, D.C., 2003. Representative hillslope methods for applying slope conditions, including the changes in soil and vegetation as the WEPP model with DEMs and GIS. Trans. ASAE 46, 1041–1049. well as convergence of the slope. The extensive database of WEPP Cooper, S.V., Neiman, K.E., Roberts, D.W., 1991. Forest Habitat Types of Northern Idaho: A Second Approximation. Gen. Tech. Rep. INT-236, USDA For. Serv., on soil, vegetation, and management of crop-, range- and forest- Intermountain Res. Stn., Boise, ID, 143 pp. land allows more broad applications. Existing grid-based hydrol- Covert, S.A., 2003. Accuracy Assessment of WEPP-based Erosion Models on Three ogy and erosion models, such as the DHSVM (Wigmosta et al., Small, Harvested and Burned Forest Watersheds. MS Thesis, Univ. Idaho, Moscow, ID. 1994; Doten et al., 2006) and MIKE-SHE (Refsgaard and Storm, Covert, S.A., Robichaud, P.R., Elliot, W.J., Link, T.E., 2005. Evaluation of runoff 1995), also have the ability to simulate both infiltration- and satu- prediction from WEPP-based erosion models for harvested and burned forest ration-excess runoff as well as water erosion. They apply finite- watersheds. Trans. ASAE 48, 1091–1100. 54 S. Dun et al. / Journal of Hydrology 366 (2009) 46–54
Dissmeyer, G.E. (Ed.), 2000. Drinking Water from Forests and Grasslands: A Janssensm, I.A., Jarvis, P., Jensen, N.O., Katul, G., Mahli, Y., Matteucci, G., Meyers, Synthesis of the Scientific Literature. Gen. Tech. Rep. SRS-39, USDA For. Serv. T., Monson, R., Munger, W., Oechel, W., Olson, R., Pilegaard, K., Paw, U.K.T., S. Res. Stn., Asheville, NC, 246 pp. Thorgeirsson, H., Valentini, R., Verma, S., Vesala, T., Wilson, K., Wofsy, S., 2002. Domenico, P.A., Schwartz, F.W., 1998. Physical and Chemical Hydrogeology, second Environmental controls over carbon dioxide and water vapor exchange of ed. John Wiley & Sons, Inc., New York. pp. 38–41. terrestrial vegetation. Agric. For. Meteorol. 113, 97–120. Doten, C.O., Bowling, L.C., Maurer, E.P., Lanini, J.S., Lettenmaier, D.P., 2006. A Lott, N., Ross, D., Sittel, M., 1997. The Winter of 96–97 West Coast Flooding. NOAA- spatially distributed model for the dynamic prediction of sediment erosion and NCDC Rep. 97-01.